| With the development and application of the technologies of the internet of things,cloud computing and the fifth generation mobile networks,the mass data storage and processing have been one of the technology focus in the field of storage.The NAND flash memories,the nonvolatile storages,are applied to phones personal computers,data centers and so on,because of the advantage of the fast reading and writing speed,small size and high capacity.However,when the NAND flash technologies help to increase the capacity and speed,the reliability of the data storage suffers from a great risk.So how to ensure the reliability of the data storage of NAND flash memory is one of the research hotspots in the field of storage.Presently,although the Low-Density Parity-Check(LDPC)code can significantly improve the reliability of the data storage in NAND flash memories,it is more and more difficult to ensure the reliability of data storage using a single error-correcting code technology for the increasing of the interference of flash channel.Therefore,it is necessary to study the technologies to further improve the reliability of the data storage,such as channel estimation and signal processing.This paper executes the analyses and simulations of the high-density multistage NAND flash memory channel,and studies a neural-network-assisted error correction algorithm and a channel parameter estimation algorithm,according to the characteristics of the threshold voltage distribution and channel noises.The research and innovation in this paper is summarized as follows:(1)The internal structure characteristics,the mechanism of the erasing and programming and the noise generation,the influence of the noises on the threshold distribution are studied in this paper.Then,we formulate the flash channel simulation model and study the channel estimation and error correction algorithm for NAND flash memory.(2)We study the LDPC code and its decoding algorithm and simulate the error performance using the formulated flash channel model.(3)According to the generation mechanism of the retention noise,we propose a flash channel estimation algorithm.Since the flash channel model can be approximated by Gaussian mixture model,the proposed algorithm minimize the square Euclidean distance between the probability of the measured threshold voltage and the Gaussian mixture model.to estimate the mean and standard deviation of the threshold voltage distribution.(4)Based on the channel characteristics of the flash memory,a relative logarithmic likelihood ratio(LLR)calculation and a neural-network-assisted error correction algorithm are proposed.The calculation of the relative LLR is to use the relative position,instead of the absolute position,of the reading reference voltage to estimate the LLR value,which makes the estimated value of LLR more stable.In addition,the stable LLR value can greatly improve the learning accuracy of the neural network in the NAND flash channel.Therefore,the neural network model trained by the relative LLR is able to correct partial error bits of LDPC code,which improves the reliability of the NAND flash memory. |
